
Some tips /things to note:
-
This dataset is rather large (over 500 images), however the extreme stylized proportions may not play well with some LORAs, especially ones that have a style baked in- you'll need to experiment a bit with weights to get the optimal result!
-
This LORA was trained on images of both highly articulated and highly unarticulated dolls, so multiple posing options (while limited in order to maintain the effect), are possible.
-
A trigger word is not necessarily needed, however adding "bratz doll" and /or "doll" to the prompt helps guide the image to the intended result.
-
Trying to generate males is extremely difficult; while there are some male photos in the dataset, there really weren't enough to accurately train off of.
-
The intended style is meant to be photorealistic, mimicking real photos of real Bratz dolls. However, in some circumstances (mostly while using LORAs with a deep anime style baked in at a high weight), I've found it possible to get a drawn look, similar to the Bratz concept art featured on doll boxes. However, this is not the intended use for this LORA and as such, results may vary heavily.
-
After playing around with numerous Pony merges, I've found CrystalPony to give the most believable results (and is the merge I used for all showcase pictures), however I definitely encourage you to try and play around with some others!
描述:
训练词语:
名称: bratzdoll.safetensors
大小 (KB): 223113
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success